1. Field of the Invention
This invention relates generally to the measurement of an alerting or other signaling tone. More particularly, it relates to a versatile method and apparatus capable of accurately measuring the duration of a tone burst, even in an environment including speech signals.
2. Background of Related Art
While the present invention relates to tone detection in general, it and its background are described with respect to a particular embodiment useful for detecting alerting tones relating to the reception of call related information, e.g., Caller ID information.
As is well known, Caller ID services permit a telephone customer to learn the identity of, or at least the originating telephone number of, a party seeking to place a telephone call to the customer. An older service referred to as Call-Waiting (CW) alerts a customer engaged in a telephone connection to another party, that a third party desires to place a telephone call to the customer. A newer service offered in telephone systems combines caller identification and call-waiting services, to not only alert a customer during a telephone connection that a third party desires to establish a connection, but also transmits to the CPE, caller identification information identifying the telephone number and/or name of the third party. This latter service is sometimes referred to as CIDCW (caller ID and call waiting) or Caller ID Type II.
Participation in a system with Caller ID and Call Waiting requires not only the presence of a central office capable of providing such service, but also CPE capable of receiving and transmitting the necessary control, acknowledgment and data signals. The general sequence of events during a telephone call in which the customer has Caller ID and Call Waiting service is as follows.
When a customer is currently engaged in a telephone conversation with a second party, a third party desiring to complete a telephone call with the customer may dial the customer""s number. During the customer""s conversation with the second party, the central office sends a subscriber alerting signal (SAS) alerting the customer that a third party desires to make a connection with the customer. The subscriber alerting signal is typically a single tone 400-hertz signal of about 500 milliseconds in duration, that is audible to the customer.
The central office follows the transmission of the subscriber alerting signal with the transmission of a CPE alerting signal (CAS). Similar to a dual tone multi-frequency signal, the CPE alerting signal has 2130 hertz and 2750 hertz components and optimally lasts for 80 milliseconds. Upon detection of the CPE alerting signal, the CPE sends a CAS acknowledgment signal (CAS ACK) to the central office, signaling to the central office that the CPE is ready to receive Caller ID information regarding the third party.
At the time the CPE transmits the CAS acknowledgment signal, it also disables the voice transmission transducer in the CPE. While the central office maintains the connection with the second party, it suppresses the CAS acknowledgment signal from the transmission sent to the second party. When the central office receives the CAS acknowledgment signal, it sends the caller identification information to the customer in the form of a frequency shift keyed (FSK) signal. The CPE can then display the caller identification information on a display unit as is known in the art. Full communication is established with the second party after receipt of the caller identification information. The customer, through the CPE can then decide how to handle the third party, such as placing the second party on hold and establishing the connection with the third party, for example.
Because of the proximity of the CPE alerting signal to voice frequencies transmitted and received in typical telephone systems, reliable detection of the CPE alerting signal can sometimes be a problem in conventional CPEs. For instance, conventional tone detectors used in this capacity sometimes experience xe2x80x9ctalkoffxe2x80x9d episodes in which CPE alerting signals are falsely detected, and xe2x80x9ctalkdownxe2x80x9d episodes in which actual CPE alerting signals are missed due to interference with speech. This erroneous operation is often the result of harmonic components of speech signals occurring in the frequency ranges of a tone signal such as the standard CPE alerting signal.
One conventional approach to eliminating talkoff and talkdown episodes during CPE alerting signal detection is described in U.S. Pat. No. 5,263,084 issued Nov. 16, 1993 to Chaput et al. The Chaput approach appears to be as follows. When the CPE receives the subscriber alerting signal, it mutes its handset for a predetermined period of time to reduce or eliminate background noise, in expectation of the occurrence of a CPE alerting signal conforming to known standards. The muting interval must be chosen to be of sufficient duration to reliably allow silence for CPE alerting signals of a given tolerance (i.e., beginning time, duration). Unfortunately, this technique requires a prior knowledge of the occurrence of an incoming tone signal. Moreover, the muting interval is typically long enough to be noticed by and cause an annoyance to customers engaged in a telephone conversation.
A different conventional approach employs the use of brute-force digital signal processing (DSP) methods, such as the use of fast Fourier transforms (FFTs), to continuously analyze each frame of a digitized signal for the presence of any/all expected tones.
Many people have made telephone calls, enjoying the convenience of communications using a telephone or other customer premises equipment. Moreover, everyone has heard some type of alerting tone, call progress tone (CPT) or other signaling tone from a central office, be it a busy tone, a dial tone, or other single or dual frequency tone used to signal customer premises equipment or a central office over a telephone line.
Generally, alerting or other signaling tones are either continuous, periodic, intermittent, or single pulse only. The duration of any particular tone pulse may be very short, e.g., 40 milliseconds (mS) or less. Generally, a minimum length of a particular tone pulse is conventionally established to provide a conventional tone detection algorithm sufficient time to detect the presence of the tone.
FIG. 4 shows a conventional tone detector 902 adapted to detect the presence of a particular tone (i.e., a particular alerting or other signaling tone) in an input signal typically received over a telephone line, and output either a xe2x80x9ctone detectedxe2x80x9d or a xe2x80x9ctone not detectedxe2x80x9d signal. Many tone detectors contain a plurality of algorithms or parameters to accommodate the detection of a plurality of expected alerting tone or other tone signals. Some signaling tones require detection of more than one frequency component, e.g., Dual Tone Multiple Frequency (DTMF) signals.
Tone detectors (e.g., DTMF dialer and call progress tone detectors) are common in most customer premises equipment. Early customer premises equipment included analog filters for detecting a tone, but more recent customer premises equipment includes a processor with digital tone detection software. Digital tone detectors usually include an algorithm implementing a discrete Fourier transform (DFT).
A DFT processes digital samples of the input signal on a frame-by-frame basis. The DFT processes each sample frame of a given length to detect the presence of a particular frequency component in the input signal to a resolution corresponding to the length of the data frame. If a particular frequency component above a predetermined energy threshold is detected in the processed frame having the proper duration, then a xe2x80x9ctone detectedxe2x80x9d or similar signal is output.
FIG. 5 shows an exemplary sampling and processing of an input signal on a frame-by-frame basis through a DFT.
In particular, a signal including a tone pulse 1002 shown in waveform (a) of FIG. 5 is input to a DFT adapted to detect the presence of the tone pulse 1002. In a worst case scenario, the tone pulse 1002 may be split fairly evenly between separate frames 907, 909 for processing by the DFT, as represented by the positioning of a standard DFT frame signal 911 shown in waveform (b) of FIG. 5 with respect to the timing of the tone pulse 1002 in the input signal. Because the tone pulse 1002 may be split between frames 907, 909 of the DFT, the DFT must be capable of detecting the tone pulse 1002 in any one frame using a shorter duration (e.g., half) of the tone pulse 1002 than is available in the input signal.
DFTs with short data frames have wide main lobes and thus include a significant amount of noise along with the desired signal, while longer data frames tend to be better at noise reduction but decrease the resolution with which a duration of a tone can be detected. Conventionally, a balance is reached between the desirous and undesirous characteristics of short and long DFT data frames in view of the particular application, and a frame length for DFT processing is chosen. Longer DFTs require more memory (e.g., ROM memory), and typically take longer to compute.
Alternatively, overlapping of frames has been implemented to increase an effective resolution, but adds significant complexity to a system design.
Accordingly, some robustness may be lost in conventional DFTs, e.g., due to the selected length of the DFT data frame, and/or due to the inevitable asynchronicity between the tone pulse 1002 and the frame boundaries of the conventional DFT. However, in many conventional applications this is not a problem, e.g., because the tone pulses are detected in relatively noiseless environments. In such a situation, a tone can be detected relatively quickly with good confidence.
Typically, there was little reason to measure the duration of a tone pulse to any high degree of accuracy. As shown in the example of FIG. 5, measurement of the duration of the tone pulse 1002 using conventional techniques has at best the resolution of the DFT framing. For instance, the presence of the tone pulse 1002 may be output by the DFT at the conclusion of frames 907, 909, and will not be cleared until a subsequent frame 917, which is significantly longer than the actual duration of the tone pulse 1002 as shown in FIG. 5. Accordingly, the measured duration of the tone burst 1002 using conventional tone detectors found in most customer premises equipment or central offices may be fairly inaccurate with respect to the actual duration of the tone pulse 1002.
Most conventional tone detectors detect the presence of tones when there is no speech on the telephone line, and thus most tone detectors need not be too robust to provide reliable results. Thus, short tones (i.e., tones having a minimum duration of, e.g., 40 mS) can be used for signaling. Moreover, in such systems, tone frequencies were not mixed with voice signals. More recently, it has become necessary to detect alerting and/or other tones in the presence of voice signals on a telephone line, and to measure their frequency with a high degree of accuracy.
FIG. 6 shows the approximate long-term average spectral energy density for continuous speech, showing that high relative spectral energy is present in near-end speech, which is in the same general range as many signaling tones used by central offices. Thus, the presence of speech together with an alerting or other tone may cause noise in the particular spectral regions being measured for the presence of a particular alerting or other signaling tone.
For instance, Call Subscriber Services such as Caller ID/Call Waiting (CIDCW). Caller ID/Call Waiting is a Type II Caller ID service allowing the transmission of call related information such as a telephone number and/or household or business name to a called party regarding a third party caller to an established telephone call between the called party and someone else. Robust tone detectors are necessary to avoid the possibility of errors such as talkoff (false tone detects) and/or talkdowns (missed tone detects).
It is estimated that for every 10 mS of increased accuracy in the measured duration of a tone pulse, talkoff errors decrease by a factor of three (3). However, as described herein above, conventional tone detectors are generally limited in the accuracy of tone pulse measurement to the resolution of the DFT frame length.
Accordingly, there is a need for apparatus and techniques to accurately and efficiently measure the duration of tone bursts to increase the robustness of tone detectors, particularly with the increase in the need to detect alerting or other signaling tones on a telephone line including speech signals or other distracting and/or distorting signals, e.g., as in Call Subscriber Services such as CIDCW.
In accordance with the principles of the present invention, apparatus for measuring a duration of a tone burst in an input signal comprises a long frame length discrete Fourier transform adapted to process long DFT data frames corresponding to the input signal. A short frame length discrete Fourier transform is adapted to process short DFT data frames corresponding to the input signal. The short DFT data frames are shorter than the long DFT data frames. The short frame length discrete Fourier transform interrupts a process cycle of the long frame length discrete Fourier transform corresponding to a long DFT data frame when a tone is initially detected by the short frame length discrete Fourier transform.
A method of detecting a tone burst in an input signal in accordance with the principles of the present invention comprises firstly processing data frames corresponding to the input signal having a long length to detect a substantial presence of a particular frequency. The data frames are also secondly processed using data frames having a short length to detect a substantial presence of the particular frequency. The step of firstly processing is halted before a current long length data frame has been fully processed if a corresponding short length data frame is detected to not include the substantial presence of the particular frequency.